-A A +A

CS2013-Strawman-IS-Intelligent Systems (DEPRECATED)

10 posts / 0 new
Last post
CS2013-Strawman-IS-Intelligent Systems (DEPRECATED)

The CS2013 Strawman Report comment period is now closed. Please see the CS2013 Ironman report for the latest CS2013 draft to comment on. Forum to comment on "IS-Intelligent Systems" Knowledge Area in the CS2013 Strawman report. We ask that comments related to specific text in the report please specify the page number and line number(s) of the text being commented on. Line numbers are provided on the far left-hand side of the each page.

Argumentation in IS Knowledge Area
Your rating: None


The list of topics described in the Intelligent Systems (IS) elective
Advanced Representation and Reasoning omits explicit mention of Argumentation.

In view of the number of existing programs within which this field features either as
an elective module in itself or as a major sub-topic of advanced KR and Reasoning provision, we
consider this to be a serious oversight and, in the following, present the main
reasons why Argumentation in AI defines a significant area that merits
consideration within an Advanced KR and Reasoning module. This case
is structured by giving a short summary reason followed by a supporting
case and evidence.

1. Preparation for subsequent advanced Masters and doctoral level research

Computational models of argumentation forms one of the most active current research areas within AI: IJCAI (2009, 2011),
AAAI (2007 onwards), KR (2010) and AAMAS (2010, 2011) all had specialist sessions within the technical program with
Argumentation (or Argumentation and Negotiation) in the heading. In addition
a dedicated international conference (COMMA) directly addressing computational
models of argumentation has become well established since its inaugural
meeting in 2006 as well as a journal (Argument and Computation, first volume 2010) specializing in this area.
Examination of publication metrics in journals such as
Artificial Intelligence (widely regarded as the leading venue for disseminating
work in AI) indicate significant interest and activity in this area, e.g. of the 25 listed most-cited
articles (since 2007) as of April 2012, 10 of these are concerned with argumentation.
Within any program there will be a number of students who become sufficiently
inspired to wish to pursue further research in the field and, given the level of current research
interest, it seems likely that computational aspects of argumentation will continue to offer a rich vein
of research issues for some time. In this regard familiarity with the basic core of work that has been accomplished
will leave such students far better prepared to make further contributions.

2. Perspective and insight into Practical Reasoning and Critical Thinking

Argumentation provides a natural platform for reasoning on many different levels so obviating any need to
distinguish between object-level, meta-level, meta-meta level etc. Hence argumentation is a natural framework for studying practical
reasoning. Systematic research on argumentation can be traced back to Toulmin (S. Toulmin, The Uses of Argument, Cambridge Univ. Press, 1959).
In that sense, argumentation is not simply a sub-field of non-monotonic reasoning nor a discipline that first emerged from it even though research on
non-monotonic reasoning and logic programming has had a great impact on its computational formalization.
In contrast, current research on argumentation provides in many aspects deeper insights
into the nature of practical reasoning than non-monotonic reasoning alone.

3. Relevance to other core areas of Computer Science

Although the historical roots of abstract argumentation are often perceived as deriving from studies and work on
non-monotonic reasoning (e.g. as outlined in the survey article of Chesnevar, Maguitman and Loui; ACM Comp. Surveys, 32:337-383, 2000) over the past decade it has increasingly been seen and applied as an important paradigm with
applications as diverse as human-computer interaction, communication and dialogue in multi-agent systems, decision support, e-democracy, AI and Law, AI and Medicine, social networks, etc.: that is domains in which practical reasoning - the analysis of what actions to perform and their supporting cases - plays a more crucial role, than simply establishing which assertions do or do not hold true.
Thus coverage of argumentation within an advanced KR elective provides additional
opportunities for the valuable pedagogical purpose of illustrating interaction between sub-fields within CS as a whole
rather than these being simply isolated unconnected specialisms.

4. Multidisciplinary Nature

While one would expect the focus of argumentation within a CS course to be primarily its
computational aspects, argumentation theory has important connections with many fields outside CS, most
notably: game theory, linguistics, philosophy, rhetoric, law, logic, politics and economics. It thus provides an ideal environment in which to promote the importance of computational ideas in very broad contexts.

5. Relevance to "real-world" applications and settings

The potential for exploiting argumentation as a computational paradigm has been widely recognised
in a number of large-scale projects from the early 1990s onwards.
Among important developments of this type are medical decision support systems under the aegis of
the Imperial Cancer Research Fund (subsequently Cancer Research UK); the use of
argumentation for international military crisis management for the US Federal Government (Project Genoa) under the auspices of DARPA. Argumentation based tools for e-democracy feature as a significant element of projects currently supported through the European Union (e.g. the IMPACT project dealing with development and evaluation of innovative prototype tools for supporting open, inclusive and transparent deliberations about public policy), an agency whose input had been instrumental in the development of the argumentation platform ASPIC.


As has been noted by Gordon (Foundations of Argumentation Technology – Habilitation Thesis; TU Berlin, 2009)

Argumentation technology uses and builds upon information technology. The concepts of information and communication
are broad enough to cover all the elements of argumentation processes, such as the assertion of a claim,
the asking of a question or the putting forward of an argument.

So that the view of argumentation as a paradigm central to informatics in general and intelligent systems in particular is very much in keeping with the the sentiments expressed, for example, in texts such as Russel and Norvig (Artificial Intelligence: A Modern Approach, 2nd ed. Prentice Hall Series in Artificial Intelligence. Prentice Hall, 2003) in recognising the importance of social, deliberative, and negotiation elements in modeling intelligent agents.

Paul E. Dunne (U. of Liverpool, UK)

with the agreement of and additional input from

Leila Amgoud (IRIT Toulouse, France)
Katie Atkinson (U. of Liverpool, UK)
Pietro Baroni (U. Brescia, Italy)
Gerd Brewka (U. Leipzig, Germany)
Sylvie Doutre (IRIT Toulouse, France)
Phan Minh Dung (Asian Institute of Technologies, Thailand)
Dov Gabbay (Kings College, London, UK)
Tom Gordon (U. of Potsdam, Germany)
Peter McBurney (Kings College, London, UK)
Pierre Marquis (Univ. d'Artois, Lens, France)
Sanjay Modgil (Kings College, London, UK)
Nir Oren (U. of Aberdeen, UK)
Henry Prakken (U. of Utrecht/U. of Groningen, Netherlands)
Chris Reed (U. of Dundee, UK)
Guillermo Simari (U. Nacional del Sur, Argentina)
Francesca Toni (Imperial College, London, UK)
Bart Verheij (U. of Groningen, Netherlands)
Doug Walton (U. of Windsor, Ontario, Canada)
Stefan Woltran (Tech. Universitat Vienna, Austria)

A Kumar
Thanks for making the case for argumentation
Your rating: None

The committee has now listed argumentation under advanced knowledge representation and reasoning - this will appear in the next version of the document. Thank you for making a persuasive case for inclusion of argumentation.

tneller's picture
"IS-Intelligent Systems" Knowledge Area Comments
Your rating: None

32-34: Use consistent capitalization of initial words only. (There may be other capitalization inconsistencies elsewhere - I have not checked thoroughly.)

55, 72: Game-tree search can and should be removed from basic search. As a game AI researcher, this is one of my favorite topics. Yet I feel it detracts from a thorough coverage of the core search problems in the rest of the topics list. Given only a 4 hour allotment, time is better spent going deeper with the other topics. Game-tree search is already properly included in "IS/Advanced Search".

56, 70: I would replace "satisfaction" with "satisfaction/optimization". I could see instructors alternatively teaching constraint optimization (subsuming constraint satisfaction) with an emphasis on different, important problems that draw more from less-commonly-taught techniques (e.g. stochastic local search).

70: Remove the hyphen between constraint and satisfaction.

77, 78, 83, 84: Given that we merely review predicate logic (77) and go in depth in core concepts using propositional logic (78) should the application learning outcomes (83 and 84) have the words "predicate" and "quantified" removed?

88: My main concern with this section is that a student presented with "Basic Machine Learning" as outlined here would likely come away with the lesson that "ML = classification". I approve of the choice of classification as a simple and important focus problem for introduction. However, there are basic exploration-versus-exploitation ML problems (e.g. n-armed bandit) in reinforcement learning. RL is indeed included in "IS/Advanced Machine Learning". Perhaps an introductory bullet that overviews different ML problems and subareas before delving into a pure focus on classification would suffice to help students better grasp the breadth and diversity of work in ML.

108-110: Lines 109 "Simulated annealing" and 110 "Genetic algorithms" are both properly subcategories of 108 "Stochastic search". Perhaps these bullets should then either be indented, or removed and instead listed on a revised line 108: "Stochastic search (including simulated annealing and genetic algorithms)".

Overall, I'm excited to see the inclusion of many new and interesting topics throughout the document. Many thanks to my excellent strawman draft contributor colleagues!

tneller's picture
MCTS Postscript
Your rating: None

between 112 and 113: Insert new line: "Monte Carlo tree search". MCTS is one of the dominant algorithms used in modern game-tree search. Techniques in this family are responsible for the significant leap forward in Computer Go performance, and one will generally find that MCTS is applied more often that minimax w/ alpha-beta or proof number search for some of the more hotly-contested games in the annual Computer Olympiad. A survey of papers in the proceedings of the collocated 2010 International Conference on Computers and Games had 14 papers primarily concerned with game-tree search. Of these, 11 were concerned with various forms of MCTS, and 3 were primarily concerned with proof-number search. There was a similar domination of MCTS papers at the collocated 2011 Advances in Computer Games.

A Kumar
Lots of good feedback
Your rating: None

Todd, thank you for providing detailed feedback on the document, the committee appreciates your effort and enthusiasm. We have incorporated most of the comments into the post-strawman document (including the comment on MCTS below). There were a couple of issues where we thought clarification was in order:
55, 72: Only Minimax is included in Basic search; Alpha-beta pruning is in advanced search. Minimax exposes students to the added complications of adversarial search and typically takes 10-20 minutes to cover in class. So, the committee felt, it should be left in Basic search.
56, 70: Constraint satisfaction is often used as an AI approach to solving problems. Constraint optimization on the other hand is a much larger topic that is typically covered in Algorithms. The committee felt that it would confuse the reader to present these two concurrently as alternatives.

Some ML should be in Core-Tier1
Your rating: None

I work in a completely different field, but it has become clear over the past decade that the statistical machine learning technologies have a power that is impacting all areas of computing research and practice. I would think that every CS graduate, without exception, needs to know what these techniques can do (though not all need to know how to use ML themselves). Thus I would elevate at least 1 hour of IS/Basic Machine Learning to Core-Tier1: I think a key outcome here is "Know examples of impacts of ML technologies in practice" and "Identify tasks that could be effected by ML techniques, and propose sources of training data for these tasks"

A Kumar
Tier I versus Tier II
Your rating: None

Thank you for arguing for the elevation of Machine Learning to Tier I status. The committee agrees with your assessment of the importance of Machine Learning in Computer Science curriculum today. This explains why it is listed as a Tier 2 core rather than an elective. Computer Science programs are expected to cover at least 90% of Tier 2 core topics in their curricula.

Additional comments (received via email)
Your rating: None

Some of the topic lists could be reduced in size by abstracting combining concepts through abstraction.

Comments from CRA Snowbird 2012 meeting
Your rating: None

[Comments below are from panel session on CS2013 are CRA Snowbird 2012]

Perhaps some amount of machine learning should be Tier1 given how large a role it has come to play and the way it allows CS to bridge to other fields.

Log in or register to post comments